Data Mining Techniques to Build A Recommender System
Academic Article in Scopus
-
- Overview
-
- Identity
-
- Additional document info
-
- View All
-
Overview
abstract
-
© 2021 IEEE.Recommender systems are not a new topic but they are still current due to their broad applications and the impact that they can have in sales or revenues. There are different approaches to build a system that makes recommendations, the selection of a method depends on several factors. For example, the available data to extract knowledge and the technical resources of the company. In this article, we focus on the use of data mining techniques to build a recommender system. Then, we describe in more detail two methods: communities finding and market basket analysis. Both methods are easy to implement and they are efficient. We present our methodology to implement them as well as the advantages and disadvantages that we found in them. We used a publicly available dataset to test both methods offline. We were interested in evaluating both methods without a ground truth reference because the model selection is one of the challenges when building a new system. For the communities finding method, we used fitness functions that allowed us to compare and select a set of communities. For the market basket analysis, we tested different values for the parameters that can be controlled: support, confidence, and the number of items in transactions. With both methods, we obtained easy to interpret results that can be used to implement a recommender system. The tested methods are suggested when there is only historical data about transactions. When more information is available, there are more robust methods that can be implemented. For example, content-based methods.
status
publication date
Identity
Digital Object Identifier (DOI)
Additional document info
has global citation frequency
start page
end page